[RISCV] Split sched classes for vrgather.vv and vrgatherei16.vv (#92768)

michaelmaitland · web-flow · commit 89245b6cf808 · 2024-05-23T12:10:46.000-04:00
These can behave different on a subtarget since EEW=16 and EMUL =
(16/SEW)*LMUL for the indices in vs1.
diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoV.td b/llvm/lib/Target/RISCV/RISCVInstrInfoV.td
@@ -1680,8 +1680,9 @@ let Predicates = [HasVInstructions] in {
 let Constraints = "@earlyclobber $vd", RVVConstraint = Vrgather in {
 defm VRGATHER_V : VGTR_IV_V_X_I<"vrgather", 0b001100>;
 def VRGATHEREI16_VV : VALUVV<0b001110, OPIVV, "vrgatherei16.vv">,
-                      SchedBinaryMC<"WriteVRGatherVV", "ReadVRGatherVV_data",
-                                    "ReadVRGatherVV_index">;
+                      SchedBinaryMC<"WriteVRGatherEI16VV",
+                                    "ReadVRGatherEI16VV_data",
+                                    "ReadVRGatherEI16VV_index">;
 } // Constraints = "@earlyclobber $vd", RVVConstraint = Vrgather
 
 // Vector Compress Instruction
diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoVPseudos.td b/llvm/lib/Target/RISCV/RISCVInstrInfoVPseudos.td
@@ -2264,8 +2264,8 @@ multiclass VPseudoVGTR_VV_EEW<int eew, string Constraint = ""> {
           defm _VV
               : VPseudoBinaryEmul<m.vrclass, m.vrclass, emul.vrclass, m, emul,
                                   Constraint, e>,
-                SchedBinary<"WriteVRGatherVV", "ReadVRGatherVV_data",
-                            "ReadVRGatherVV_index", mx, e, forceMergeOpRead=true>;
+                SchedBinary<"WriteVRGatherEI16VV", "ReadVRGatherEI16VV_data",
+                            "ReadVRGatherEI16VV_index", mx, e, forceMergeOpRead=true>;
         }
       }
     }
diff --git a/llvm/lib/Target/RISCV/RISCVSchedSiFive7.td b/llvm/lib/Target/RISCV/RISCVSchedSiFive7.td
@@ -928,6 +928,7 @@ foreach mx = SchedMxList in {
     defvar IsWorstCase = SiFive7IsWorstCaseMXSEW<mx, sew, SchedMxList>.c;
     let Latency = !add(Cycles, 3), AcquireAtCycles = [0, 1], ReleaseAtCycles = [1, !add(1, Cycles)] in {
       defm "" : LMULSEWWriteResMXSEW<"WriteVRGatherVV", [SiFive7VCQ, SiFive7VA], mx, sew, IsWorstCase>;
+      defm "" : LMULSEWWriteResMXSEW<"WriteVRGatherEI16VV", [SiFive7VCQ, SiFive7VA], mx, sew, IsWorstCase>;
       defm "" : LMULSEWWriteResMXSEW<"WriteVCompressV", [SiFive7VCQ, SiFive7VA], mx, sew, IsWorstCase>;
     }
   }
@@ -1273,6 +1274,8 @@ defm "" : LMULReadAdvance<"ReadVFSlideV", 0>;
 defm "" : LMULReadAdvance<"ReadVFSlideF", 0>;
 defm "" : LMULSEWReadAdvance<"ReadVRGatherVV_data", 0>;
 defm "" : LMULSEWReadAdvance<"ReadVRGatherVV_index", 0>;
+defm "" : LMULSEWReadAdvance<"ReadVRGatherEI16VV_data", 0>;
+defm "" : LMULSEWReadAdvance<"ReadVRGatherEI16VV_index", 0>;
 defm "" : LMULReadAdvance<"ReadVRGatherVX_data", 0>;
 defm "" : LMULReadAdvance<"ReadVRGatherVX_index", 0>;
 defm "" : LMULReadAdvance<"ReadVRGatherVI_data", 0>;
diff --git a/llvm/lib/Target/RISCV/RISCVSchedSiFiveP600.td b/llvm/lib/Target/RISCV/RISCVSchedSiFiveP600.td
@@ -716,6 +716,7 @@ foreach mx = ["MF8", "MF4", "MF2", "M1"] in {
     defvar IsWorstCase = SiFiveP600IsWorstCaseMX<mx, SchedMxList>.c;
     let Latency = 3, ReleaseAtCycles = [1] in {
       defm "" : LMULSEWWriteResMXSEW<"WriteVRGatherVV", [SiFiveP600VEXQ1], mx, sew, IsWorstCase>;
+      defm "" : LMULSEWWriteResMXSEW<"WriteVRGatherEI16VV", [SiFiveP600VEXQ1], mx, sew, IsWorstCase>;
       defm "" : LMULSEWWriteResMXSEW<"WriteVCompressV", [SiFiveP600VEXQ1], mx, sew, IsWorstCase>;
     }
   }
@@ -736,6 +737,7 @@ foreach mx = ["M2", "M4", "M8"] in {
     defvar IsWorstCase = SiFiveP600IsWorstCaseMXSEW<mx, sew, SchedMxList>.c;
     let Latency = 6, ReleaseAtCycles = [LMulLat] in {
       defm "" : LMULSEWWriteResMXSEW<"WriteVRGatherVV", [SiFiveP600VEXQ1], mx, sew, IsWorstCase>;
+      defm "" : LMULSEWWriteResMXSEW<"WriteVRGatherEI16VV", [SiFiveP600VEXQ1], mx, sew, IsWorstCase>;
       defm "" : LMULSEWWriteResMXSEW<"WriteVCompressV", [SiFiveP600VEXQ1], mx, sew, IsWorstCase>;
     }
   }
@@ -1071,6 +1073,8 @@ defm "" : LMULReadAdvance<"ReadVFSlideV", 0>;
 defm "" : LMULReadAdvance<"ReadVFSlideF", 0>;
 defm "" : LMULSEWReadAdvance<"ReadVRGatherVV_data", 0>;
 defm "" : LMULSEWReadAdvance<"ReadVRGatherVV_index", 0>;
+defm "" : LMULSEWReadAdvance<"ReadVRGatherEI16VV_data", 0>;
+defm "" : LMULSEWReadAdvance<"ReadVRGatherEI16VV_index", 0>;
 defm "" : LMULReadAdvance<"ReadVRGatherVX_data", 0>;
 defm "" : LMULReadAdvance<"ReadVRGatherVX_index", 0>;
 defm "" : LMULReadAdvance<"ReadVRGatherVI_data", 0>;
diff --git a/llvm/lib/Target/RISCV/RISCVScheduleV.td b/llvm/lib/Target/RISCV/RISCVScheduleV.td
@@ -521,6 +521,7 @@ defm "" : LMULSchedWrites<"WriteVISlide1X">;
 defm "" : LMULSchedWrites<"WriteVFSlide1F">;
 // 16.4. Vector Register Gather Instructions
 defm "" : LMULSEWSchedWrites<"WriteVRGatherVV">;
+defm "" : LMULSEWSchedWrites<"WriteVRGatherEI16VV">;
 defm "" : LMULSchedWrites<"WriteVRGatherVX">;
 defm "" : LMULSchedWrites<"WriteVRGatherVI">;
 // 16.5. Vector Compress Instruction
@@ -749,6 +750,8 @@ defm "" : LMULSchedReads<"ReadVFSlideF">;
 // 16.4. Vector Register Gather Instructions
 defm "" : LMULSEWSchedReads<"ReadVRGatherVV_data">;
 defm "" : LMULSEWSchedReads<"ReadVRGatherVV_index">;
+defm "" : LMULSEWSchedReads<"ReadVRGatherEI16VV_data">;
+defm "" : LMULSEWSchedReads<"ReadVRGatherEI16VV_index">;
 defm "" : LMULSchedReads<"ReadVRGatherVX_data">;
 defm "" : LMULSchedReads<"ReadVRGatherVX_index">;
 defm "" : LMULSchedReads<"ReadVRGatherVI_data">;
@@ -956,6 +959,7 @@ defm "" : LMULWriteRes<"WriteVSlideI", []>;
 defm "" : LMULWriteRes<"WriteVISlide1X", []>;
 defm "" : LMULWriteRes<"WriteVFSlide1F", []>;
 defm "" : LMULSEWWriteRes<"WriteVRGatherVV", []>;
+defm "" : LMULSEWWriteRes<"WriteVRGatherEI16VV", []>;
 defm "" : LMULWriteRes<"WriteVRGatherVX", []>;
 defm "" : LMULWriteRes<"WriteVRGatherVI", []>;
 defm "" : LMULSEWWriteRes<"WriteVCompressV", []>;
@@ -1120,6 +1124,8 @@ defm "" : LMULReadAdvance<"ReadVFSlideV", 0>;
 defm "" : LMULReadAdvance<"ReadVFSlideF", 0>;
 defm "" : LMULSEWReadAdvance<"ReadVRGatherVV_data", 0>;
 defm "" : LMULSEWReadAdvance<"ReadVRGatherVV_index", 0>;
+defm "" : LMULSEWReadAdvance<"ReadVRGatherEI16VV_data", 0>;
+defm "" : LMULSEWReadAdvance<"ReadVRGatherEI16VV_index", 0>;
 defm "" : LMULReadAdvance<"ReadVRGatherVX_data", 0>;
 defm "" : LMULReadAdvance<"ReadVRGatherVX_index", 0>;
 defm "" : LMULReadAdvance<"ReadVRGatherVI_data", 0>;

Original file line number	Diff line number	Diff line change
`@@ -2264,8 +2264,8 @@ multiclass VPseudoVGTR_VV_EEW<int eew, string Constraint = ""> {`
`2264`	`2264`	`defm _VV`
`2265`	`2265`	`: VPseudoBinaryEmul<m.vrclass, m.vrclass, emul.vrclass, m, emul,`
`2266`	`2266`	`Constraint, e>,`
`2267`		`- SchedBinary<"WriteVRGatherVV", "ReadVRGatherVV_data",`
`2268`		`- "ReadVRGatherVV_index", mx, e, forceMergeOpRead=true>;`
	`2267`	`+ SchedBinary<"WriteVRGatherEI16VV", "ReadVRGatherEI16VV_data",`
	`2268`	`+ "ReadVRGatherEI16VV_index", mx, e, forceMergeOpRead=true>;`
`2269`	`2269`	`}`
`2270`	`2270`	`}`
`2271`	`2271`	`}`
Original file line number	Diff line number	Diff line change
`@@ -716,6 +716,7 @@ foreach mx = ["MF8", "MF4", "MF2", "M1"] in {`
`716`	`716`	`defvar IsWorstCase = SiFiveP600IsWorstCaseMX<mx, SchedMxList>.c;`
`717`	`717`	`let Latency = 3, ReleaseAtCycles = [1] in {`
`718`	`718`	`defm "" : LMULSEWWriteResMXSEW<"WriteVRGatherVV", [SiFiveP600VEXQ1], mx, sew, IsWorstCase>;`
	`719`	`+ defm "" : LMULSEWWriteResMXSEW<"WriteVRGatherEI16VV", [SiFiveP600VEXQ1], mx, sew, IsWorstCase>;`
`719`	`720`	`defm "" : LMULSEWWriteResMXSEW<"WriteVCompressV", [SiFiveP600VEXQ1], mx, sew, IsWorstCase>;`
`720`	`721`	`}`
`721`	`722`	`}`
`@@ -736,6 +737,7 @@ foreach mx = ["M2", "M4", "M8"] in {`
`736`	`737`	`defvar IsWorstCase = SiFiveP600IsWorstCaseMXSEW<mx, sew, SchedMxList>.c;`
`737`	`738`	`let Latency = 6, ReleaseAtCycles = [LMulLat] in {`
`738`	`739`	`defm "" : LMULSEWWriteResMXSEW<"WriteVRGatherVV", [SiFiveP600VEXQ1], mx, sew, IsWorstCase>;`
	`740`	`+ defm "" : LMULSEWWriteResMXSEW<"WriteVRGatherEI16VV", [SiFiveP600VEXQ1], mx, sew, IsWorstCase>;`
`739`	`741`	`defm "" : LMULSEWWriteResMXSEW<"WriteVCompressV", [SiFiveP600VEXQ1], mx, sew, IsWorstCase>;`
`740`	`742`	`}`
`741`	`743`	`}`
`@@ -1071,6 +1073,8 @@ defm "" : LMULReadAdvance<"ReadVFSlideV", 0>;`
`1071`	`1073`	`defm "" : LMULReadAdvance<"ReadVFSlideF", 0>;`
`1072`	`1074`	`defm "" : LMULSEWReadAdvance<"ReadVRGatherVV_data", 0>;`
`1073`	`1075`	`defm "" : LMULSEWReadAdvance<"ReadVRGatherVV_index", 0>;`
	`1076`	`+defm "" : LMULSEWReadAdvance<"ReadVRGatherEI16VV_data", 0>;`
	`1077`	`+defm "" : LMULSEWReadAdvance<"ReadVRGatherEI16VV_index", 0>;`
`1074`	`1078`	`defm "" : LMULReadAdvance<"ReadVRGatherVX_data", 0>;`
`1075`	`1079`	`defm "" : LMULReadAdvance<"ReadVRGatherVX_index", 0>;`
`1076`	`1080`	`defm "" : LMULReadAdvance<"ReadVRGatherVI_data", 0>;`